rhashtable: Use __vmalloc with GFP_ATOMIC for table allocation
authorHerbert Xu <herbert@gondor.apana.org.au>
Fri, 4 Dec 2015 14:39:56 +0000 (22:39 +0800)
committerDavid S. Miller <davem@davemloft.net>
Fri, 4 Dec 2015 21:53:05 +0000 (16:53 -0500)
commitd3716f18a7d841565c930efde30737a3557eee69
tree5f0599a1e442a490a1d87e610110f9245825bc2d
parent6a61d4dbf4f54b5683e0f1e58d873cecca7cb977
rhashtable: Use __vmalloc with GFP_ATOMIC for table allocation

When an rhashtable user pounds rhashtable hard with back-to-back
insertions we may end up growing the table in GFP_ATOMIC context.
Unfortunately when the table reaches a certain size this often
fails because we don't have enough physically contiguous pages
to hold the new table.

Eric Dumazet suggested (and in fact wrote this patch) using
__vmalloc instead which can be used in GFP_ATOMIC context.

Reported-by: Phil Sutter <phil@nwl.cc>
Suggested-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: David S. Miller <davem@davemloft.net>
lib/rhashtable.c